A tool called Pangram flagged 100% artificial intelligence in the text.
That number was enough to put Thélyson Orélien at the center of one of the biggest controversies in French literature in recent memory. 🎭
The Canadian-Haitian writer, 38 years old, had just won the Prix du Roman Fnac, the first major literary prize of the season in France, with his debut novel C’était ça ou mourir, which roughly translates to It Was That or Die. And as if the prize weren’t enough, Orélien is also a finalist for two of the most prestigious awards in the French language: the Goncourt and the Renaudot, with results expected soon. The book became an almost instant sensation, and the Canadian publisher behind it went so far as to call it the biggest seller in 60 years — a feat any debut author would dream of celebrating for a long time.
But before the celebration could last too long, a little-known group called Balance ton Claude, which loosely translates to something like Expose Your Claude — a direct reference to Anthropic’s AI assistant — dropped a bomb in the middle of the literary party. The accusation, posted on the social media platform X on a Tuesday, was blunt: the novel had allegedly been written, almost entirely, by an artificial intelligence. According to the group, tests were run on multiple passages of the manuscript and all returned the same result: 100% AI. Orélien denied everything immediately. And that kicked off a debate that goes far beyond a single book or a single author. 🤔
After all, where does human authorship end when AI tools are increasingly present in everyday creative work? And more importantly: can we actually trust these AI detectors to decide the fate of a literary career? That’s exactly what we’re going to dig into here.
What is Pangram and why is it at the center of this story
Pangram is an AI-generated content detection tool designed to identify textual patterns that supposedly indicate the use of language models in writing. The group Balance ton Claude described the tool as by far the most accurate on the market, claiming the chance of it making a mistake was nearly insignificant. To back up their argument, group members even cited that Pangram would only deliver a wrong assessment once every 24,400 analyses of AI-generated texts. Impressive on paper, but the technical reality is a lot more complicated than that number suggests.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
Tools like this work by analyzing statistical features of a text — things like the predictability of word choices, sentence structure, and the absence of variations that naturally show up in human writing. Worth noting that the more modern detectors no longer compare text against text to check for plagiarism, the way older tools used to. Instead, they analyze probabilities and statistics to try to determine whether certain linguistic patterns give away the use of AI. The problem is that this kind of analysis is far from an exact science, and experts themselves warn that false positive rates can be significant, especially when the text being analyzed is a translation or has gone through intensive editorial revisions.
In Thélyson Orélien’s case, Pangram returned a score of 100% AI, which sounds alarming at first glance. But specialists in computational linguistics and natural language processing have been warning for some time that this kind of extreme score can paradoxically be a sign that the tool isn’t properly calibrated for a certain type of text. Very formal texts, heavily edited works, or writing in languages other than English — like French — tend to produce skewed scores in tools primarily trained on English-language data. What should have been a piece of technical evidence ended up becoming fuel for an unprecedented public accusation in the French literary scene.
It’s also worth remembering that Pangram, like other similar tools on the market, has no access to the author’s creative process. It analyzes the final product, without considering drafts, notes, revisions, or any other evidence of the human work behind the text. This creates a dangerous scenario where a number generated by an algorithm can be used to destroy a reputation built over years, without a proper chance for a technical defense. The controversy around Orélien exposed that vulnerability in a very concrete way. 🧐
Thélyson Orélien and the journey before the storm
Thélyson Orélien was born in 1988 in the city of Gonaives, in northern Haiti. Before leaving his home country, his early works were mainly composed of poetry. In 2010, he moved to Montreal, Canada, where he worked as a journalist before diving fully into a literary career. The novel C’était ça ou mourir marks his debut in the genre, and the work has been described by critics and readers as a dense, emotionally charged narrative with a distinct literary voice — exactly the kind of quality that tends to catch the eye of literary prize juries.
The story in the book carries an obvious emotional weight. It follows a teacher who flees gang violence in Haiti and lives through a series of adventures before finally making it to Canada, where he rediscovers the right to breathe without fear. It’s a narrative that speaks directly to the author’s own life journey, which makes the accusations even more delicate. The Prix du Roman Fnac is awarded at the start of the French literary season and carries considerable weight in the publishing world, capable of catapulting an unknown author into the spotlight of public and media attention. The success was so significant that the book is expected to be translated into around 20 languages.
Facing the accusations, Orélien responded firmly to the newspaper Libération. He stated that he has always loved creation and literature, and that he saw no reason whatsoever to feel compelled to use a machine. The author also pointed out that his traditions are Haitian and Caribbean, and that the images, repetitions, and rhythms present in the text are the living word — something deeply personal that belongs only to him. According to his statements, he was putting together a dossier with his editor to prove that he wrote the book authentically. An important detail: Orélien claimed the first draft of the work was written in 2019, well before AI text-generation programs became widely available to the public.
The publisher Grasset also came to the author’s defense, calling the allegations atrociously defamatory against a writer with an exceptional life story, whose creative work was original and powerful. Even so, in an era where a screenshot travels faster than any denial, reputational damage can happen before the truth has a chance to catch up. This phenomenon, which we know all too well from the world of fake news, has now hit the world of French literature with full force. 📚
Are AI detectors reliable enough to accuse someone?
This is perhaps the most important question in this entire story. Artificial intelligence detection tools for text are still at a pretty experimental stage, and that’s not an isolated opinion — it’s a growing consensus among researchers in the field. Critics of Pangram and other detectors point out that these tools are unreliable, since different programs often deliver conflicting results when analyzing the exact same text. Research shows that these systems can incorrectly flag human writing as AI-generated and, at the same time, miss texts that were actually produced by machines, especially when they’ve been edited or paraphrased.
The structural problem with these tools lies in how they were trained. The truth is that these detectors don’t actually know who or what wrote a given text. Instead, they look for patterns that tend to appear in AI-generated writing, but the crucial detail is that those same patterns can also show up naturally in work written by humans. On top of that, many of these programs are built around standardized language use — usually English or formal French — and because of that, they have real difficulty correctly interpreting texts that use dialects or linguistic variations from places like the Caribbean, which is exactly the case with Orélien’s writing.
This means that an author who writes very carefully, with well-crafted sentences and precise vocabulary, or who carries specific cultural markers in their style, can end up being flagged as AI simply because their text falls outside the pattern the tool learned to classify as human. It’s a cruel irony: the more original and unique the writer, the greater the risk of being confused with a machine. 😅
In the context of the controversy involving Orélien, this raises a very serious ethical question. Using the result of a tool like Pangram as the basis for a public accusation, without accounting for the technology’s known technical limitations, is irresponsible at best. The group Balance ton Claude, whose members remain unknown, acted as if an algorithm’s output were equivalent to concrete proof of fraud — which is far from the case. French literature, which has always had a passionate relationship with intellectual debate, now finds itself forced to confront a question that the tech world still hasn’t resolved: how to distinguish, fairly and reliably, what is human from what is machine, especially when the two can coexist within the same creative process.
A case that already has a similar precedent
Interestingly, this isn’t the first time a literary award has found itself in the middle of a storm involving suspicions of AI use. Orélien’s episode is quite reminiscent of the controversy surrounding the Trinidad and Tobago writer Jamir Nazir and his work The Serpent in the Grove. The Commonwealth Short Story Prize was actually withdrawn from him in July amid allegations that the text had been produced with generative AI. The story, however, had a different ending than expected: the prize was eventually returned to the author after the organizers declared themselves satisfied that the writing was genuine.
This precedent matters because it shows how accusations of this kind can be premature and, in many cases, reversed when examined with more patience and care. It also reinforces just how unprepared the literary market still is to deal with this new reality. With AI becoming increasingly normalized, the fight over C’était ça ou mourir will only feed the debate further about what mix of human and artificial creativity is considered acceptable. After all, it’s entirely possible that a writer generates text with AI and then adapts it, or takes the opposite path — writing an original text and then asking an AI to review it. Where exactly to draw the ethical line remains an open question.
The bigger debate this episode opens up
The case of Thélyson Orélien isn’t just about a book or about a prize. It touches on something that’s becoming increasingly urgent as artificial intelligence moves deeper into creative processes: the need to establish clear, fair, and technically grounded criteria for evaluating authorship in the 21st century. Publishers, literary prizes, and academies everywhere are being forced to create policies around AI use in creation, often without the technical knowledge needed to make those decisions responsibly. The result is a set of vague rules that can both protect and harm authors in unpredictable ways.
There’s also a very important human dimension to this discussion. Authors living in situations of economic or social vulnerability — immigrants, writers from developing countries, or newcomers without a solid support network — are the most exposed when accusations like this circulate without a solid foundation. A literary reputation is a fragile asset, built over years and capable of being eroded in hours. When the tool making the accusation is an algorithm with known limitations, and the group publishing the accusation acts without formal accountability, the environment becomes especially hostile for those still trying to establish themselves in the publishing world.
This episode also raises a reflection on how society will handle the coexistence of human creation and AI tools in everyday creative work. It’s common for writers to use advanced grammar checkers, writing assistants, and other technologies without anyone considering it dishonest. The line between using a support tool and outsourcing creation to an AI still isn’t well defined, and until that debate moves forward in a mature way, cases like Orélien’s will keep coming up. French literature, with its deep tradition of critical thinking, has a unique opportunity to lead this conversation responsibly. ✍️
What’s clear, at the end of the day, is that AI detection technology isn’t ready to serve as a courtroom. And until that changes, any accusation based solely on an algorithm’s score should be treated with the same caution as any other inconclusive evidence.
